Abstract

Harmful algal blooms (HABs) and eutrophication became one of the most important problems in global environmental issues that has a grievous threat to freshwater habitats, biodiversity, drinking water security, and socio-economic stability. The methods of traditional in-situ sampling and in-laboratory analysis are also valid, but have a limited scope of their usefulness due to their high labour-intensive nature and the lack of real-time or large-scale analyses. Current developments in satellite-based Earth observation systems and the usage of deep learning algorithms have now offered the benefit of high-resolution, scalable, and rapid monitoring of aquatic systems. This research paper compiles a client remote sensing system based on the deep learning methodology to identify, measure, and predict the dynamics of eutrophication, and HAB growth on the basis of the multispectral and hyperspectral images of the Sentinel-2, Landsat-8/9, MODIS, and PRISMA satellites. The suggested system will use convoluted neural networks (CNNs), long short-term memory (LSTM) networks, the Vision Transformers (ViTs) systems and a combination of the CNN/LSTM systems that can achieve the learning of spectral-spatial representations and spatial features and temporal evolving of the blooms respectively. The most important water quality indicators, such as chlorophyll- a (Chl-a) concentration, turbidity, total suspended solids and nitrogen- phosphorus proxies are estimated with the help of regression and classification models that are trained on harmonised satellite data and field-measured ground truth. The experimental outcomes on several freshwater lakes and reservoirs show that the hybrid deep learning model has more than 94% classification accuracy on the level of the bloom intensity, and a root-mean-square error (RMSE) of Chl-a prediction is less than 7 percent, which is better than conventional machine learning baselines. The framework is also capable of 3- to 7-day predictions of the behaviour of blossoms, which could greatly benefit the early-warning and resource management systems. This research can contribute greatly to remote sensing-met water quality monitoring and interventions through offering an operationally versatile, cost-effective and scalable solution to the increasing effects of eutrophication and HAB events, providing effective decision-support tools to environmental agency, population health departments and freshwater resource managers in the US and beyond.